Sign in with
Sign up | Sign in
Your question
Solved

Poor performance on video transcoding (Vegas 12 and other video apps) with an AMD 7970Ghz Edition

Last response: in Graphics & Displays
Share
April 28, 2014 4:43:25 PM

To keep a long story short I made the switch from a nVidia GTX560 1GB to an AMD 7970Ghz Edition 6GB after comparing benchmarks on several websites.

I'm a casual gamer and video editor, I shoot videos with a DSLR and a helmet camera and very often I need to convert them to save space on the hardware and share online.

Before I switched the video cards (3 days ago) I benchmarked Vegas 12, Aiseesoft Video Converter, CineBench R15, Luxmark, etc. Except for Luxmark which gives me twice the score everything else is slower than the GTX 560.

There's an MP4 file that I've converted using Aiseeoft Video Converter with CUDA support that is 1:12 long that I converted to H.264 (Youtube profile) and it took just 13 seconds with CUDA, the same video with the same settings takes 33 seconds with AMD APP Acceleration instead, Aieseesoft supports both CUDA and AMD APP Acceleration.

On SONY Vegas 12 the rendering speed with CUDA on H.264 was honorable, with OpenCL on the AMD 7970 it is dreadful.

I can go on and on but will stop here and look ahead to hear from other fellows that use similar video editing/converting apps what are their thoughts on AMD's tecnology, is it not supported yet or not it doesn't perform as well as the CUDA does on lower end video cards?

Thanks
April 28, 2014 4:51:52 PM

Right now AMD's openCL is simply not the standard and thus suffers from not having near as much time for optimizations to make it in the programs. Simply put as it stands right now CUDA is the standard and OpenCL is an afterthought.

There are a few programs that are optimized better for openCL and a google search should reveal them.
m
0
l
April 28, 2014 4:58:00 PM

Spectre694 said:
Right now AMD's openCL is simply not the standard and thus suffers from not having near as much time for optimizations to make it in the programs. Simply put as it stands right now CUDA is the standard and OpenCL is an afterthought.

There are a few programs that are optimized better for openCL and a google search should reveal them.


I thought OpenCL was supposed to be an "open" standard to avoid proprietary technology such as CUDA, in that case if an app supports OpenCL it should run fine on any video card make/model that supports the OpenCL implementation, if that's not the case then the users are basically screwed again.

Adobe claims it supports AMD acceleration via OpenCL on it's latest products however AdobeMedia Encoder is installed on my machine and didn't get any benefit from the new graphics card.

The whole thing with AMD just seems broken, I'm starting to feel bad for not picking a nVidia GTX770 until I'm able to extract the juice from this red beast.
m
0
l

Best solution

April 28, 2014 5:19:22 PM

Paul F said:
Spectre694 said:
Right now AMD's openCL is simply not the standard and thus suffers from not having near as much time for optimizations to make it in the programs. Simply put as it stands right now CUDA is the standard and OpenCL is an afterthought.

There are a few programs that are optimized better for openCL and a google search should reveal them.


I thought OpenCL was supposed to be an "open" standard to avoid proprietary technology such as CUDA, in that case if an app supports OpenCL it should run fine on any video card make/model that supports the OpenCL implementation, if that's not the case then the users are basically screwed again.

Adobe claims it supports AMD acceleration via OpenCL on it's latest products however AdobeMedia Encoder is installed on my machine and didn't get any benefit from the new graphics card.

The whole thing with AMD just seems broken, I'm starting to feel bad for not picking a nVidia GTX770 until I'm able to extract the juice from this red beast.


It is the "open" standard but Nvidia has put something like 7 or so years of money, marketing and development behind CUDA. versus openCL which has has some nonprofit group running it and only recently (last 2 years maybe) had any real backers like AMD (even then kinda halfheartedly until more recently) It is essentially the same reason that DirectX is the dominant gaming API vs OpenGl.
Share
April 28, 2014 5:55:57 PM

It's really sad since I bought a 280X to render faster.>.> Anyway, use Sony AVC since it kinda works. It lessens the load of cpu but the load on gpu is only between 10%-15%.:/
m
0
l
April 29, 2014 8:14:49 AM

karlsanada13 said:
It's really sad since I bought a 280X to render faster.>.> Anyway, use Sony AVC since it kinda works. It lessens the load of cpu but the load on gpu is only between 10%-15%.:/


Here are my findings on my setup:

Input:
Format: MPEG-4 AVC
Length: 3:29
Bitrate: 5000bps

With nVidia GTX560: (Monitored using Zotac Firestorm)
GPU Usage: 40% <-> 50%
CPU Usage: 80% <-> 90%
Elapsed Time: 2:20

With AMD 7970 (Monitored using AMD System Monitor)
GPU Usage: 17% <-> 21%
GPU Clock: 500MHz
CPU Usage: 45% <-> 55%
Elapsed Time: 3:20

Setup:
-> Asus P6X58D-E
-> Core i7 930 @ 2.8Ghz
-> 12GB Patriot Viper Triple Channel
m
0
l
April 29, 2014 8:18:55 AM

Interesting finding, adding a LineDrawing effect from NewBlue increases the rendering time in around 20-30 seconds and these values get changed:

GPU Usage: 38% <-> 41%
GPU Clock: 1050MHz
CPU Usage: Doesn't change, still around 50%
m
0
l
!